Biography
I am a senior research scientist at Meta working on Responsible Generative AI. Prior to joining Meta I was a part of Language and Information Technologies (LIT) team at Microsoft Research. I am generally interested in artificial intelligence and creating reliable and fair natural language and information technologies.
In the past few years, my research has been focused on Responsible Large Language Models, building Natural Language Understanding (NLU) systems and tackling out-of-distribution challenges in these systems. Some of these challenges are domain adaptations, compositional generalization, and transferring societal biases in multi-lingual transfer techniques and language models.
Before joining Microsoft, I was working towards my Ph.D. at the University of Washington, Seattle. The title of my dissertation is “Learning in Complex Dynamic Systems: with Applications to Perpetual Flight, Energy Management, Distributed Decision Making, and Social Networks.” During my PhD program, I developed distributed optimization algorithms and machine learning methods and applied these algorithms on classification problems in large-scale data sets, social network analysis, and autonomous multi-agent systems.
Selected Publications
Refer to Google Scholar for full list
2022
2021
Meghana Moorthy Bhat, Saghar Hosseini, Ahmed H. Awadallah, Paul Bennett, Weisheng LiSay `YES’ to Positivity: Detecting Toxic Language in Workplace Communications
EMNLP 2021 | November 2021
View Publication
2020
Jieyu Zhao, Subhabrata (Subho) Mukherjee, Saghar Hosseini, Kai-Wei Chang, Ahmed H. Awadallah Ahmed Elgohary, Saghar Hosseini, Ahmed H. AwadallahGender Bias in Multilingual Embeddings and Cross-Lingual Transfer
ACL 2020 | July 2020
View PublicationSpeak to your parser: Interactive Text-to-SQL with natural language feedback
ACL 2020 | July 2020
View Publication | Download
2019
Petar Stojanov, Ahmed H. Awadallah, Paul Bennett, Saghar Hosseini Wei Wang, Saghar Hosseini, Ahmed H. Awadallah, Paul Bennett, Chris QuirkOn Domain Transfer When Predicting Intent in Text
NeurIPS Workshop on Document Intelligence | December 2019
View PublicationContext-Aware Intent Identification in Email Conversations
SIGIR 2019 | July 2019
View Publication
Education
Ph.D.
2011-2016University of Washington, Seattle, WA
During my Ph.D. I developed new distributed machine learning algorithms with the application for autonomous multi-agent coordination, social networks, and large-scale text classification. I received the Amelia Earhart fellowship award from Zonta International Foundation twice. This fellowship is awarded to 35 fellows around the globe each year who are pursuing advanced studies in the aerospace-related sciences. Moreover, I received the Graduate School Top Scholars Award from the University of Washington and the best session presentation award from the American Control conference.
Master of Science
2009-2011University of California, Irvine, CA
I received my master's degree in Aerospace engineering and during my studies, I received three graduate fellowship awards from the University of California and one award from the Society of Women Engineers.
Bachelor of Science
2004-2009Sharif University of Technology, Tehran, Iran
I received my bachelor's degree in Aerospace engineering and during my studies, I was a member of Sharif University’s swimming team.
Recent Projects
Fairness in Language Models
Large-scale pre-trained language models (PLMs) such as BERT and GPT have recently achieved great success in varieties of Natural Language Processing (NLP) tasks due to their architectures and a huge number of model parameters. These large-scale PLMs capture knowledge from massively labeled and unlabeled data which could contain latent societal biases. Development on top of PLMs without appropriately considering these inherent biases, risks exacerbating and perpetuating societal biases when these models are deployed.In this project, we are focused on societal biases in different NLP models such as propagation of gender bias in multi-lingual word embeddings and PLMs.
Natural Language Interfaces to Databases
Interfaces to large-scale databases have become ubiquitous by enabling open platforms that any developer can interact with. Currently, there are about 24,000 Web APIs available1for a wide range of domains including productivity, social media, payment, shopping, education, messaging, science, games, and maps. In addition to WEb API, a large amount of data is available on SQL databases. This opens an interesting avenue for improving users’ productivity through human-machine interfaces.In this project, we focus on leveraging users’ feedback to improve the interfaces. Moreover, we study the out-of-distribution challenges such as multi-lingual transfer and compositional generalization.
My Outdoor
Adventures
In my spare time, I'm somewhere in the mountains.
Error: No feed found.
Please go to the Instagram Feed settings page to create a feed.